Distributed Applications
Networking Basics
Week 6
What is a “Network?”



Depends on what level you’re at
One person’s “network” is another person’s “application”
OSI Seven Layer Model
FTP, HTTP,
7. Application




The physical wire itself
Ethernet, 802.11b
Routing protocols
...
SMTP, etc.
6. Presentation
5. Session
4. Transport
TCP
3. Network
IP
2. Data Link
ARP, RARP
1. Physical
Ethernet
For Our Purposes: The Internet




We’re application programmers
In terms of OSI, we’re defining/using
our own application-layer protocol
Sits atop TCP/IP, the lingua franca
of the Internet
For almost every networked
application you will ever want to
build, this will be the lowest layer
in the stack you’ll need to care
about
7. Application
FTP, HTTP,
SMTP, etc.
6. Presentation
5. Session
4. Transport
TCP
3. Network
IP
2. Data Link
ARP, RARP
1. Physical
Ethernet
Topology of the Internet
gatech.edu
My Home Network
google.com
Some Terminology: Protocols

Protocols: rules that facilitate information exchange among
programs on a network



Example from human world: “roger” and “over” for radio geeks
Similar to how you design the interfaces between objects in your
program

A callback expects to get a certain set of parameters in a certain order

You need to know this in order to use the callback
Likewise:


A networked program expects you to communicate with it in certain
ways (using certain messages, in a known format)
You need to know this in order to use the program
Some Terminology: Servers

Server: a (generally) long-lived program that sits around waiting
for connections to it


Examples: web server, mail server, file server, IM server
“Server” implies that it does something useful (delivers a service)




Web server: provides access to HTML documents
Mail server: allows retrieval, sending, organization of email messages
File server: provides remote access to files and directories
IM server: provides info about online users, passes messages between
them
Some Terminology: Clients

Client: a program that connects to a server to use whatever
service it provides

Examples:




Web browser connects to web servers to access/view HTML
documents
Mail client (Outlook, etc.) connects to mail servers for mail storage,
transmission
IM clients connect to IM servers to access info about who is on, etc.
Most servers can be connected to by multiple clients at the same
time
Some Terminology: Host


Host: Simply a machine that’s connected to the network
Generally running clients and/or servers

The machine “hosts” a server
The Next Phase of the Project

We’ll be building the networking part of the IM program


For the IM assignment:


Enhancing the GUI code to talk to an either an IM server on the networking
I’ll provide a sample IM server, and documentation on its protocol
Important concept: understanding a protocol specification

Useful for when you want to write a program that talks to an existing server (and thus
has its own existing, documented protocol)

Side concept: designing your own protocols


We’ll talk about this, but won’t do it for the project (unless you want to go nuts and get all
fancy...)
Should give you experience in using basic Internet-style networking, debugging, etc.
What Will You Have to Do?
1. Connect to the other machine(s)
 Know how to refer to it: which machine do you want to connect to?

Know how to perform the connection
Know how to deal with errors (server is down, etc.)
2. Send messages to it (e.g., “I’m online now!”)


Know how to “marshall” arguments
Know how to do the transmission

Know how to deal with errors (server crashed while sending, etc.)

3. Receive messages from it (e.g., list of online users)
 Know how to “unmarshall” arguments

Know how to read data

Know how to deal with errors (e.g., got unexpected data from server, etc.)
4. Disconnect from it
 This is the easy part!
Why All the Focus on Errors?


Networking in inherently error-prone
Different than single application programming


Networking: errors may be caused by reasons outside of your
control





Errors generally result from a bug, and just crash entire program
Network is down, server has crashed, server slow to respond, etc.
During a chat I could shut my laptop and walk away
Someone could trip over the power cord for an access point
Networks can’t even guarantee that messages will get from A to B
Good goal: robustness


Your program should survive the crash of another program on the
network, receiving malformed data, etc
“Defensive programming”
Networking 101
Internet Addressing


Every machine on the Internet has an address
Internet addresses are sequences of 4 bytes



Addresses identify a particular machine on the Internet


Usually written in “dotted quad” notation
Examples: 192.168.13.40, 13.2.117.14
Example: 64.223.161.104 is the machine www.google.com
One special address

127.0.0.1

localhost
Refers to the local machine always

Where do IP Addresses Come
From?

You can’t just set your IP address to any random value and have it
work



In most cases a service called DHCP will take care of this for you





The rest of the Internet won’t know how to reach you
You have to use values that are compatible with whatever network
you’re on
Dynamic Host Configuration Protocol
Assigns you a valid IP address when you boot your machine, wake your
laptop, etc.
E.g., LAWN at Georgia Tech
IP address may change from time to time: in other words, don’t count
on this being your address forever
If DHCP isn’t available, you may have to set your IP address by
hand, but only with a value provided by an administrator
Why Do You Need to Know
This?

First off: don’t change your IP address for this class!


Second: if you get an address from DHCP (which you probably do),
you can’t count on having this address forever


You can only do harm!
So don’t hard-code it into any programs
Third: if you want to debug clients and servers on the same
machine, you can use the localhost address

But don’t hardcode this either, since it would keep you from working
when client and server are on different machines
Public Versus Private Addressing


Not all IP addresses may be reachable from any given machine
Simple case: machines behind a firewall


Example: my old machine at PARC was 13.1.0.128, but only reachable
from within PARC
More complex case:


Some IP addresses are private (also called non-routable)
Three blocks of addresses that cannot be connected to from the larger
Internet



10.0.0.0 - 10.255.255.255
172.16.0.0 - 172.31.255.255
192.168.0.1 - 192.168.255.255
Why Private Addresses?

Two reasons: IP address conservation and security

Public addresses uniquely define a given machine


Private addresses can be reused (although not on the same network)


There’s a limited number of these, and they’re running out
Probably hundreds of thousands of machines with 192.168.0.1 on
private networks (corporation internal, homes, etc.)
Certain network configs let you share a single public IP address across
multiple private machines


Network Address Translation
Built into most home routers

E.g., BellSouth gives me the address 68.211.58.142

My router gives my home machines 192.168 addresses

Connections out are translated so that it looks like they come from 68.211.58.142

Internal machines are “invisible” since they have non-routed addresses
Why Do You Need to Know
This?

Servers running on machines with private IP addresses are not
reachable from machines not on that network




Ok if you’re running your client and service on the same network
Ok if you’re running your client and service on the same machine
Not ok if, e.g., your server is at home and you client is at Georgia Tech
Aside: this is the reason that many people pay for an extra “static” IP
address at home--so that they can run servers that have a fixed IP
address that is visible throughout the Internet
Naming


When you go to a web browser, you don’t type in 64.223.161.104,
you type in www.google.com
The Domain Name Service




A big distributed database of all the machines on the Internet
Each organization manages its own little portion of it
Maps from host names to IP addresses
Ultimately, the Internet runs on IP addresses. Names are a
convenience for humans


When you type www.google.com, the browser resolves that name to an
IP address by talking to a DNS server
If name resolution can’t be done (DNS is down; you’re not connected to
the network), then browsing will fail
Naming Configuration

Much like IP addressing, you may not have much control over the
DNS name for your machine


In general, you won’t have a name resolvable by DNS, even if your
machine has a “local” name
In the CoC, CNS sets up DNS names for the machines they
administer, mapping them to fixed IP addresses



If you were to take these machines to different networks (where they
get different IP addresses), those names would no longer work
Resolve to the incorrect address
Personally owned machines, even if they get an IP address from DHCP,
generally get sucky names, if they get a name at all

Example: lawn-199-77-214-212 on my laptop
Why Do You Need to Know
This?



General all-around erudition and cocktail party conversation :-)
Even though we’re used to using names to refer to machines on the
Greater Internet, you’ll probably be reduced to using IP addresses
for this assignment
We may be able to run a server on a well-known machine,
administered by CNS, in which case you’d be able to specify it by
name
Ports

What if you’ve got multiple servers running on a single host?



When you tell a client to connect to a given machine, how does it
know which server running on that machine to talk to?
Ports: Let you address different servers running on the same
machine



E.g., a machine might have a web server, mail server, FTP server, ...
Think of IP addresses as the street address for an apartment building
Ports specify the individual apartments
Ports are just numbers that range from 0-65,535
More On Ports








Back to the question: when I type www.google.com into my browser...

It knows to go to 64.233.161.104

But how does it know which is the port for the google web server?
Well-known ports: certain common Internet services use standard port
numbers:

Web servers: port 80

FTP servers: port 21
Terminology: we say that the FTP server runs on port 21, meaning that this is
the port at which it is waiting for clients to connect to it
Reserved ports: ports 0-1024 reserved for privileged programs
Servers specify which port they run on when they start
Clients specify both the IP address of the desired host, and the port number,
when they connect to a server
Clients outgoing connections also have a port, but generally you don’t need to
know what it is
Only one client or service can run on a port at any given time
Why Do You Need to Know
This?


If you’re writing a client for an existing service, you’ll have to know
what port is is running on in order to connect to it
If you write a service, you’ll need to run it on a port that will be
known by its clients




Can be a fixed port number that you decide on, and tell your clients
Can let the system assign you a random one, but then you’ll need some
way to communicate this to clients
You can’t choose ports in the reserved range
Good practice is to use relatively high numbers (e.g., 5,000 50,000)
Network Programming 101
Basic Network Programming




One unified concept for dealing with the network at the Internet
layer: sockets
Basically similar across all platforms (Java, C, Python, etc.)
De facto standard (slight differences across platforms, languages)
So what’s a socket?




An endpoint for communication
May be connected to another endpoint, in another program on the net
Lets you read from it and write to it, much like a file
Adds some additional operations specific to networking
Network Programming from the
Client’s Perspective
1. Create a socket
2. Bind it to an address on a client machine

Both endpoints of a communication have addresses, including ports
3. Connect it to the server, by specifying its address and port

This call blocks until the connection is successful, or times out
4. Read and write to and from the socket, to get and send data
5. Close the socket when you’re done with it
Network Programming from the
Server’s Perspective
1. Create a socket
2. Bind it to an address on the server machine

This sets the port for the socket
3. Listen for incoming connections
4. Accept any connection that comes in.

This call blocks until a new connection comes in

This produces a new socket, paired with the client, and just for
communication with that client
This socket can be read, written, and closed independently from the
socket used for any other client



Meanwhile, original listening socket can go back to listening
Allows you to have multiple ongoing client connections at one time
5. Close the listening socket when you’re done accepting connections
Example: Basic Socket
Programming in Jython
import socket
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.connect(”192.168.2.54”, 45235)
s.listen(5)
newSock, clientAddress = s.accept()
s.send(”hello world”)
reply = sa.recv(1024)
s.close()
Writing a Simple Server
(All of this code is on the web site, as net-sampler.py)
import socket
import sys
class SimpleServer:
def __init__(self, port):
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self.sock.bind('', port)
self.sock.listen(5)
while 1:
requestSock, peerAddress = self.sock.accept()
print "Accepted connection from", peerAddress
while 1:
input = requestSock.recv(1024)
if not input:
print "Peer closed connection"
break
requestSock.send(input)
requestSock.close()
if __name__ == "__main__":
port = 7777
if len(sys.argv) > 1:
port = sys.argv[1]
server=SimpleServer(port)
Writing a Simple Client
import socket
import sys
class SimpleClient:
def __init__(self, serverAddr, serverPort):
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.connect(serverAddr, serverPort)
def sendToServer(self, message):
self.sock.send(message)
return self.sock.recv(1024)
def close(self):
self.sock.close()
if __name__ == "__main__":
if len(sys.argv) != 3:
sys.exit(1)
else:
client = SimpleClient(sys.argv[1], int(sys.argv[2]))
while 1:
string = sys.stdin.readline()
if string == "close\n":
client.close()
sys.exit(0)
else:
response = client.sendToServer(string)
print "Server replied '", response, "'"
Extra Useful Tricks

Figuring out what you’re connected to:


Figuring out your local address:


s.getpeername() returns a tuple of (address, port) indicating what
you’re connected to (or what has connected to you)
s.getsockname() returns a tuple of (address, port) indicating your local
address. Useful when you need to know what port your service is on
Making life easier:




s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
Tells the OS that it’s ok to reuse a port number
Example: you find a bug, kill your server, fix it, and restart
Without this call, OS may prevent the port from being reused until
some timeout expires
Multi-threaded Servers

Problem with previous simple server:




While it’s processing requests from one client, every other client must
queue up
Only when first client dies does the next one in the queue get handled
Bad, since most servers should support connections by multiple
clients at the same time
Common approach: multi-threaded servers


One thread to hang around waiting for clients to appear
One thread to handle each client; terminates when client is done
Multi-Threaded Server Example
import socket
import sys
import threading
class MTServer:
def __init__(self, port):
self.sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
self.sock.bind('', port)
self.sock.listen(1)
while 1:
requestSock, peerAddress = self.sock.accept()
handler = Handler(requestSock)
class Handler:
def __init__(self, requestSock):
self.requestSock = requestSock
self.thread = threading.Thread(target=self.handle)
self.thread.start()
def handle(self):
while 1:
input = self.requestSock.recv(1024)
if not input:
break
self.requestSock.send(input)
self.requestSock.close()
if __name__ == "__main__":
port = 7777
if len(sys.argv) > 1:
port = sys.argv[1]
server=MTServer(port)
Message Formatting

Any messages you send to a server must be parseable by it



Recipient must be able to decipher what you sent it
Must know when it has reached the end of the message
There are many ways of encoding messages
The Joy of ASCII

Many protocols use a simple text-based encoding

Example: HTTP
GET /index.html HTTP/1.0

Example: SMTP
HELO rutabaga.cc.gatech.edu
MAIL From: Keith Edwards <keith@cc>
DATA
Hello there!
.




Parameters and commands encoded using simple, regular format
Marshalling: the process of gathering parameters and encoding them for
transmission
Unmarshalling: the process of unpacking the received data for use by your program
Goal should be machine parseability for ease of implementation; human parseability
for ease of debugging
More Complex Data


What about very complex data?
Example: marshalling an arbitrary Jython dictionary





{”name”: “keith, “location”: (2.425, 1.783, 0.892), “info”: {”email”:
“keith@cc”, “phone”: 56783}, “buddies”: “[ralph”, “fred”, “betty”] }
You could create a string representation that is parseable and
“rebuildable” on the other end
Sometimes called flattening the dictionary to a string
Parsing at the recipient can be very difficult
Need to account for arbitrary objects that might be stored in
dictionaries (including custom-defined objects)
Is There an Easier Way?

Most “standard” services just bite the bullet and use ASCII




Perhaps with more complex formatting atop it, such as XML
ASCII--since it’s universal--lets you program a client in any language
that speaks the necessary protocol
The marshalling/unmarshalling of complicated parameters can be a
significant part of the complexity in dealing with a given service
But: If you know you’ll only be working with clients in a particular
language, you can take some short cuts
Serialization


Serialization is the process of automatically creating a representation
of complex data that can be shipped over the wire
Generally built in to the programming language itself



Opaque: with most of these systems, you don’t care what the onthe-wire representation is



So: can work with custom-defined data types without special work by
the programmer
Present in Java, Python, Jython, ...
Generally complex; generally non-ASCII
System takes care of the chores of generating it, and parsing it
Terminology: a serialization system is one approach to simplifying
the marshalling and unmarshalling of arguments
Serialization in Jython/Python

Serialization provided by the pickle library


You “pickle” objects for transmission over the wire
Works for any Jython data type, including custom-defined objects


However: some objects may “depickle” with data intact, but not behave
as expected
Classic example: swing widgets
Sending Dictionaries Using Pickle


On the sending side:
import pickle
dict = {”name”: “keith, “location”: (2.425, 1.783, 0.892), “info”: {”email”:
“keith@cc”, “phone”: 56783}, “buddies”: “[ralph”, “fred”, “betty”] }
data = pickle.dumps(dict)
s.send(data)
On the receiving side:
data = s.recv(1024)
dict = pickle.loads(data)
Combining Pickling with Other
Techniques


Pickled objects are opaque--you can’t easily parse the data yourself
Can format messages that combine ASCII with pickled objects

Have to be careful about leaving the pickled data intact

Sender:
s.send(”HELLO “ + pickle.dumps(dict))

Receiver:
data = s.recv(1024)
index = data.find(’ ‘)
command = data[0:idx]
args = data[idx+1:]

Another approach is to create a data structure that represents the entire message
and pickle it

Sender:
s.send(pickle.dumps((”Hello”, dict)))

Receiver:
pickle.loads(s.recv(1024))
Instant Messaging Assignment




Turn the GUI front end into a working network-ified program
Grab the server off the class web page
Understand the protocol it speaks
Integrate it into your client



Connect to the server
Send messages to it in response to starting up, user events (such as
new chats), etc.
Be prepared to receive messages from it


Asynchronous notifications of online users: necessitates having a thread to
listen for messages!
Responses to client-initiated messages
Getting Started


Get code off the web site: imserver.zip
 Contains newserver.py, easynet.py, timer.py
Running the server

jython newserver.py

Will run on port 6666

Generates a lot of debugging messages (don’t run under JES though)
Look at the handle messages in the server if you need to see what it’s doing


Create a client to connect to this port

Start small! Create a new file net.py

Generate a message to tell the server that you’re online

Next, make the online user list “real”: thread to listen for incoming messages

Debug by running multiple instances of the client (as different users)

Pay attention to server debugging messages!
Iron out the connection, messaging issues then integrate it

The IM Server Protocol

Uses the “command string plus pickled arguments” approach




First space in a message delineates the two
Clients announce themselves when they first start
Server periodically sends updated online user status
Clients request servers create new conversations

Tell the server to invite, specifying desired users

Server creates a conversation, giving it a unique conversation ID
Server issues invitations to all clients, indicating the conversation ID





Clients join, providing the specified conversation ID
Clients tell server to send message to parties in a conversation, by
specifying both the message and the conversation ID
Server propagates message to all members of the conversation
Clients can leave conversations by specifying their ID
The IM Server Protocol
CLIENT
SERVER
REGISTER [username, status]
ONLINE_USERS {username->status}
STATUS [status]
Authentication,
Status,
Buddies
GOODBYE
INVITE [username, username, ...]
INVITATION [convID, invitationSource]
LIST_CONVERSATIONS
CONVERSATIONS {convID, [username, username, ...]}
Chat
JOIN [convID]
SEND_MESSAGE [convID, message]
MESSAGE [convID, sender, message]
LEAVE [convID]
ERROR message
Problems
The IM Server Protocol
Clients
Server
REGISTER [username, status]
Sent when client comes online
ONLINE_USERS {username -> status}
Provides clients with the list of online users, and their status
STATUS [status]
Change client status
INVITATION [convID, invitationSource]
Sent to all invited users (including the initiating one) after an
INVITE messages
GOODBYE
Tell the server that a client is disconnecting
CONVERSATIONS {convID -> [username, username, ...]}
Response to LIST_CONVERSATIONS. Includes a dictionary
mapping from all conversation IDs to lists of users
INVITE [username, username, username, ... ]
Request the server to create a new chat with the indicated
users
MESSAGE [convID, sender, message]
Tell the client that a message has been received, indicating the
sender and the conversation
LIST_CONVERSATIONS
Request a list of all ongoing conversations
ERROR message
Tell the client that something has gone wrong
JOIN [convID]
Join the specified conversation
SEND_MESSAGE [convID, message]
Send a message to the members of the specified conversation
LEAVE [convID]
Leave the indicated conversation
47
A Few Tips

Write some utility methods for common operations

Example: receiving variable-length replies:
def receive(self, sock):
reply = “”
while 1:
dataReceived = sock.recv(1024)
if not dataReceived:
break
reply = reply + dataReceived
return reply
48
More Tips

Probably want each connection handled by its own thread
def runService(self):
while 1:
requestSock, peerAddress = self.sock.accept()
handler = Handler(requestSock)

Make a new Handler class that starts a thread that reads from requestSock, figures out
the incoming command, and dispatches it
class Handler:
def __init__(self, requestSock):
self.thread = threading.Thread(target=self.handle)
self.thread.start()
def handle(self):
input = self.requestSock.recv(1024)
if not input:
self.requestSock.close()
return
....
49